feat(dataset): 补全样本原文与删除操作

fix(dataset): 统一大小与版本元数据
2026-07-27 12:26:19 +08:00 · 2026-07-27 12:26:09 +08:00
13 changed files with 722 additions and 160 deletions
--- a/backend/app/api/v1/endpoints/platform.py
+++ b/backend/app/api/v1/endpoints/platform.py
@@ -1,7 +1,7 @@
 from __future__ import annotations

-import uuid
 import json
+import uuid
 from pathlib import Path
 from typing import Any

@@ -426,6 +426,14 @@ async def dataset_preview(file_id: str) -> dict[str, Any]:
        raise fail(404, "dataset file not found")


+@router.get("/dataset-manage/records/{file_id}/sources")
+async def dataset_record_sources(file_id: str) -> dict[str, Any]:
+    try:
+        return ok({"items": get_platform_store().dataset_file_record_sources(file_id)})
+    except KeyError:
+        raise fail(404, "dataset file not found")
+
+
@router.get("/dataset-manage/versions/{file_id}")
 async def dataset_versions(file_id: str) -> dict[str, Any]:
    try:
--- a/backend/app/db/platform_store.py
+++ b/backend/app/db/platform_store.py
@@ -1,8 +1,8 @@
 from __future__ import annotations

-import json
 import hashlib
 import hmac
+import json
 import math
 import re
 import secrets
@@ -17,7 +17,6 @@ import psycopg

 from app.core.config import get_settings

-
 ALL_PERMISSIONS = [
    "dashboard",
    "fine-tune",
@@ -61,6 +60,71 @@ def safe_float(value: Any, default: float = 0) -> float:
        return default


+_SIZE_UNIT_BYTES = {
+    "B": 1,
+    "KB": 1024,
+    "MB": 1024**2,
+    "GB": 1024**3,
+    "TB": 1024**4,
+}
+
+
+def parse_size_bytes(value: Any) -> int:
+    """把历史字符串大小统一换算为字节，供接口返回稳定的数值字段。"""
+    if isinstance(value, bool):
+        return 0
+    if isinstance(value, (int, float)):
+        return max(0, int(value))
+    match = re.fullmatch(
+        r"\s*([0-9]+(?:\.[0-9]+)?)\s*(B|KB|MB|GB|TB)?\s*",
+        str(value or ""),
+        flags=re.IGNORECASE,
+    )
+    if not match:
+        return 0
+    amount = float(match.group(1))
+    unit = (match.group(2) or "B").upper()
+    return max(0, round(amount * _SIZE_UNIT_BYTES[unit]))
+
+
+def version_number(value: Any, default: int = 0) -> int:
+    try:
+        number = int(value)
+    except (TypeError, ValueError):
+        return default
+    return number if number > 0 else default
+
+
+def dataset_file_version_summary(file_row: PgRow) -> dict[str, Any]:
+    versions = json_loads(file_row.get("versions"), [])
+    versions = versions if isinstance(versions, list) else []
+    active_version_id = str(
+        file_row.get("active_version_id")
+        or file_row.get("current_version_id")
+        or ""
+    )
+    active_version = next(
+        (
+            item
+            for item in versions
+            if isinstance(item, dict) and str(item.get("id") or "") == active_version_id
+        ),
+        None,
+    )
+    current_version_no = version_number(
+        (active_version or {}).get("version_no")
+        or (active_version or {}).get("version")
+        or file_row.get("version_no"),
+        default=1 if active_version_id or versions else 0,
+    )
+    return {
+        "active_version_id": active_version_id or None,
+        "current_version_id": active_version_id or None,
+        "current_version_no": current_version_no or None,
+        "version_count": len(versions),
+    }
+
+
 def parse_training_metric_line(line: str) -> dict[str, float] | None:
    if "loss" not in line and "learning_rate" not in line:
        return None
@@ -1099,11 +1163,30 @@ class PlatformStore:

    def _dataset(self, conn: PgConnection, row: PgRow) -> dict[str, Any]:
        files = conn.execute(
-            """SELECT id, name, size, active_version_id, create_time,
+            """SELECT id, name, size, size_bytes, active_version_id,
+                      current_version_id, version_no, versions, create_time,
                      record_count, metadata
               FROM dataset_files WHERE dataset_id=? ORDER BY create_time, id""",
            (row["id"],),
        ).fetchall()
+        decoded_files: list[dict[str, Any]] = []
+        for file_row in files:
+            metadata = json_loads(file_row.get("metadata"), {})
+            file_size_bytes = int(file_row.get("size_bytes") or 0)
+            if file_size_bytes <= 0:
+                file_size_bytes = parse_size_bytes(file_row.get("size"))
+            decoded_files.append(
+                {
+                    "id": file_row["id"],
+                    "name": file_row["name"],
+                    "size": file_row["size"],
+                    "size_bytes": file_size_bytes,
+                    **dataset_file_version_summary(file_row),
+                    "create_time": file_row["create_time"],
+                    "record_count": int(file_row.get("record_count") or 0),
+                    "split": metadata.get("file_split"),
+                }
+            )
        dataset_metadata = json_loads(row.get("metadata"), {})
        split_counts = dict(dataset_metadata.get("split_counts") or {})
        if row.get("source") == "task" and not split_counts:
@@ -1113,26 +1196,33 @@ class PlatformStore:
                (row["id"],),
            ).fetchall()
            split_counts = {str(item["split"]): int(item["count"]) for item in split_rows}
+        total_size_bytes = sum(item["size_bytes"] for item in decoded_files)
+        if not decoded_files:
+            total_size_bytes = int(row.get("size_bytes") or 0)
+            if total_size_bytes <= 0:
+                total_size_bytes = parse_size_bytes(row.get("size"))
+        current_version_nos = sorted(
+            {
+                int(item["current_version_no"])
+                for item in decoded_files
+                if item.get("current_version_no")
+            }
+        )
        return {
            **dict(row),
+            "size_bytes": total_size_bytes,
+            "current_version_no": (
+                current_version_nos[0] if len(current_version_nos) == 1 else None
+            ),
+            "current_version_nos": current_version_nos,
+            "version_count": sum(int(item["version_count"]) for item in decoded_files),
            "metadata": dataset_metadata,
            "split_counts": {
                "train": int(split_counts.get("train", 0) or 0),
                "validation": int(split_counts.get("validation", 0) or 0),
                "test": int(split_counts.get("test", 0) or 0),
            },
-            "files": [
-                {
-                    "id": f["id"],
-                    "name": f["name"],
-                    "size": f["size"],
-                    "active_version_id": f["active_version_id"],
-                    "create_time": f["create_time"],
-                    "record_count": int(f.get("record_count") or 0),
-                    "split": json_loads(f.get("metadata"), {}).get("file_split"),
-                }
-                for f in files
-            ],
+            "files": decoded_files,
        }

    def create_dataset(self, payload: dict[str, Any]) -> dict[str, Any]:
@@ -1192,12 +1282,24 @@ class PlatformStore:
        now = utcnow()
        file_id = new_id("file")
        version_id = f"{file_id}_v1"
-        size = f"{max(1, len(content.encode('utf-8')) // 1024)} KB"
+        size_bytes = len(content.encode("utf-8"))
+        size = f"{size_bytes} B"
+        record_count = len([line for line in content.splitlines() if line.strip()])
+        version = {
+            "id": version_id,
+            "version": 1,
+            "version_no": 1,
+            "create_time": now,
+            "description": "uploaded",
+            "size_bytes": size_bytes,
+            "record_count": record_count,
+        }
        conn.execute(
            """
            INSERT INTO dataset_files
-            (id, dataset_id, name, size, content, active_version_id, versions, create_time)
-            VALUES (?, ?, ?, ?, ?, ?, ?, ?)
+            (id, dataset_id, name, size, content, active_version_id, versions, create_time,
+             current_version_id, size_bytes, record_count, version_no)
+            VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, 1)
            """,
            (
                file_id,
@@ -1206,16 +1308,69 @@ class PlatformStore:
                size,
                content,
                version_id,
-                json_dumps([{"id": version_id, "version": 1, "create_time": now, "description": "uploaded"}]),
+                json_dumps([version]),
                now,
+                version_id,
+                size_bytes,
+                record_count,
            ),
        )
-        count = len([line for line in content.splitlines() if line.strip()])
        conn.execute(
-            "UPDATE datasets SET count=count+?, size=? WHERE id=?",
-            (count, size, dataset_id),
+            """UPDATE datasets
+               SET count=count+?, record_count=record_count+?,
+                   size_bytes=size_bytes+?, size=((size_bytes+?)::text || ' B')
+               WHERE id=?""",
+            (record_count, record_count, size_bytes, size_bytes, dataset_id),
        )
-        return {"id": file_id, "name": name, "size": size}
+        return {
+            "id": file_id,
+            "name": name,
+            "size": size,
+            "size_bytes": size_bytes,
+            "current_version_no": 1,
+            "version_count": 1,
+        }
+
+    def dataset_file_record_sources(self, file_id: str) -> list[dict[str, Any]]:
+        """返回发布样本关联的真实原文，供数据集详情核对生成内容。"""
+        with self.connect() as conn:
+            file_row = conn.execute(
+                "SELECT id FROM dataset_files WHERE id=?", (file_id,)
+            ).fetchone()
+            if not file_row:
+                raise KeyError(file_id)
+            rows = conn.execute(
+                """
+                SELECT records.line_no, records.instruction, records.input, records.output,
+                       COALESCE(
+                           NULLIF(preview.edited_content, ''),
+                           preview.original_content,
+                           ''
+                       ) AS source_text,
+                       CASE
+                           WHEN preview.edited_content IS NOT NULL
+                                AND preview.edited_content <> ''
+                           THEN TRUE ELSE FALSE
+                       END AS preprocessed
+                FROM dataset_records AS records
+                LEFT JOIN data_process_preview_items AS preview
+                  ON preview.id=records.preview_item_id
+                WHERE records.dataset_file_id=?
+                ORDER BY records.line_no NULLS LAST, records.created_at, records.id
+                """,
+                (file_id,),
+            ).fetchall()
+            return [
+                {
+                    "line_no": int(item.get("line_no") or index + 1),
+                    "instruction": str(item.get("instruction") or ""),
+                    "input": str(item.get("input") or ""),
+                    "output": str(item.get("output") or ""),
+                    "source_text": str(item.get("source_text") or ""),
+                    "preprocessed": bool(item.get("preprocessed")),
+                }
+                for index, item in enumerate(rows)
+            ]

    def dataset_file(self, file_id: str) -> PgRow:
        with self.connect() as conn:
--- a/backend/tests/test_platform_dataset_metadata.py
+++ b/backend/tests/test_platform_dataset_metadata.py
@@ -0,0 +1,46 @@
+from __future__ import annotations
+
+from app.db.platform_store import dataset_file_version_summary, parse_size_bytes
+
+
+def test_parse_size_bytes_supports_legacy_units() -> None:
+    assert parse_size_bytes("21563 B") == 21563
+    assert parse_size_bytes("1.5 KB") == 1536
+    assert parse_size_bytes("2 MB") == 2 * 1024**2
+    assert parse_size_bytes(4096) == 4096
+    assert parse_size_bytes("unknown") == 0
+
+
+def test_dataset_file_version_summary_uses_active_version_metadata() -> None:
+    summary = dataset_file_version_summary(
+        {
+            "active_version_id": "file-1-v3",
+            "current_version_id": "file-1-v1",
+            "version_no": 1,
+            "versions": (
+                '[{"id":"file-1-v1","version":1},'
+                '{"id":"file-1-v3","version_no":3}]'
+            ),
+        }
+    )
+
+    assert summary == {
+        "active_version_id": "file-1-v3",
+        "current_version_id": "file-1-v3",
+        "current_version_no": 3,
+        "version_count": 2,
+    }
+
+
+def test_dataset_file_version_summary_uses_normalized_version_number_as_fallback() -> None:
+    summary = dataset_file_version_summary(
+        {
+            "active_version_id": "",
+            "current_version_id": None,
+            "version_no": 1,
+            "versions": "[]",
+        }
+    )
+
+    assert summary["current_version_no"] == 1
+    assert summary["version_count"] == 0
--- a/frontend/scripts/regression-dataset-preview.mjs
+++ b/frontend/scripts/regression-dataset-preview.mjs
@@ -15,6 +15,7 @@ const apiPath = path.join(root, 'src/api/modules/dataset.ts')
 const adapterPath = path.join(root, 'src/mock/adapter.ts')
 const mockPath = path.join(root, 'src/mock/data.ts')
 const recordsPath = path.join(root, 'src/views/dataset/datasetRecords.ts')
+const recordEditorPath = path.join(root, 'src/views/dataset/preview/useDatasetRecordEditor.ts')
 const versionsPath = path.join(root, 'src/mock/datasetVersions.ts')
 const source = fs.readFileSync(previewPath, 'utf8')
 const previewLogicSource = source.slice(0, source.indexOf('<style'))
@@ -27,11 +28,12 @@ const apiSource = fs.readFileSync(apiPath, 'utf8')
 const adapterSource = fs.readFileSync(adapterPath, 'utf8')
 const mockSource = fs.readFileSync(mockPath, 'utf8')
 const recordsSource = fs.readFileSync(recordsPath, 'utf8')
+const recordEditorSource = fs.readFileSync(recordEditorPath, 'utf8')
 const versionsSource = fs.readFileSync(versionsPath, 'utf8')
 const recordsModuleCode = ts.transpileModule(recordsSource, {
  compilerOptions: { module: ts.ModuleKind.ESNext, target: ts.ScriptTarget.ES2022 },
 }).outputText
-const { parseDatasetRecords, updateDatasetRecord } = await import(
+const { deleteDatasetRecord, parseDatasetRecords, updateDatasetRecord } = await import(
  `data:text/javascript;base64,${Buffer.from(recordsModuleCode).toString('base64')}`
 )
 const versionsModuleCode = ts.transpileModule(versionsSource, {
@@ -65,6 +67,12 @@ expect(previewSurfaceSource.includes('class="record-table"'), '样本数据应
 expect(previewSurfaceSource.includes('fixed="right"'), '逐条编辑操作列应固定在表格右侧')
 expect(previewSurfaceSource.includes('v-for="fieldKey in tableFieldKeys"'), '企业表格应根据数据结构动态生成字段列')
 expect(previewSurfaceSource.includes("emit('edit', asDatasetRecord(row))"), '每条样本应向主页面发送编辑事件')
+expect(previewSurfaceSource.includes("emit('delete', asDatasetRecord(row))"), '每条样本应提供删除操作')
+expect(previewSurfaceSource.includes('label="原文参照"'), '样本列表应展示生成内容对应的原文参照')
+expect(source.includes('getDatasetFileRecordSources'), '详情页应从后端读取发布样本的真实原文')
+expect(recordEditorSource.includes('deleteDatasetRecord('), '删除样本应只修改当前文件草稿')
+expect(recordEditorSource.includes('删除后将暂存到当前草稿'), '删除样本前应说明保存版本后才生效')
+expect(source.includes('selectedFileSize'), '详情文件大小也应统一显示为 MB')
 expect(previewSurfaceSource.includes('暂存修改'), '单条编辑应先暂存到当前页面')
 expect(source.includes('hasPendingVersionChanges && isViewingActiveVersion'), '存在暂存修改时才应显示保存版本按钮')
 expect(source.includes('保存版本'), '文件工具栏应提供保存版本按钮')
@@ -85,10 +93,10 @@ expect(source.includes('hasUnsavedChanges'), '在线编辑器应跟踪未保存
 expect(source.includes('onBeforeRouteLeave'), '离开页面时应保护未保存修改')
 expect(previewSurfaceSource.includes("event.key.toLowerCase() === 's'"), '在线编辑器应支持快捷键保存')
 expect(source.includes('loadVersions(selectedFile)'), '版本内容加载失败后应支持重新加载')
-expect(source.includes('function resetRecordEditor()'), '主页面应提供统一的编辑器状态清理函数')
-const resetEditorStart = source.indexOf('function resetRecordEditor()')
-const updateFieldStart = source.indexOf('function updateEditField', resetEditorStart)
-const resetEditorSource = source.slice(resetEditorStart, updateFieldStart)
+expect(recordEditorSource.includes('function resetRecordEditor()'), '编辑器逻辑应提供统一的状态清理函数')
+const resetEditorStart = recordEditorSource.indexOf('function resetRecordEditor()')
+const updateFieldStart = recordEditorSource.indexOf('function openRecordEditor', resetEditorStart)
+const resetEditorSource = recordEditorSource.slice(resetEditorStart, updateFieldStart)
 for (const marker of [
  'editorVisible.value = false',
  'editingRecord.value = null',
@@ -141,6 +149,9 @@ const afterLines = updated.split('\n')
 expect(afterLines[0] === beforeLines[0] && afterLines[1] === '', '单条保存不得改写其他样本或空行')
 expect(JSON.parse(afterLines[2]).instruction === '第二条已修改', '单条保存应准确更新目标样本')
 expect(Array.isArray(JSON.parse(afterLines[2]).output), '单条保存不应丢失数组等非字符串字段类型')
+const deleted = deleteDatasetRecord(updated, 'samples.jsonl', 2)
+expect(!deleted.includes('第二条已修改'), '删除样本应准确移除目标 JSONL 记录')
+expect(deleted.split('\n')[0] === beforeLines[0], '删除样本不得改写其他 JSONL 记录')
 const invalid = parseDatasetRecords('{invalid json}', 'samples.jsonl')
 expect(invalid.records[0].kind === 'invalid' && invalid.records[0].error, '非法 JSON 行应被单独标记并给出错误')
 const repairedJson = updateDatasetRecord('{invalid json}', 'samples.json', 0, { repaired: true })
--- a/frontend/scripts/regression-dataset-task-tab.mjs
+++ b/frontend/scripts/regression-dataset-task-tab.mjs
@@ -42,7 +42,11 @@ assert.match(
 )
 assert.match(viewSource, /return dataList\.value\.filter\(\(item\) => item\.source !== 'task'\)/)
 assert.doesNotMatch(viewSource, /数据任务产生的数据集[\s\S]*?return \[\]/)
-assert.match(viewSource, /:search-fields="\['task_id', 'name', 'description'\]"/, '数据任务搜索应支持任务 ID')
+assert.match(viewSource, /:search-fields="\['task_id', 'name'\]"/, '数据任务搜索应支持任务 ID')
+assert.match(viewSource, /formatMegabytes\(row\.size_bytes, row\.size\)/, '数据集大小应统一转换为 MB')
+assert.match(viewSource, /<el-table-column label="版本"/, '描述列应替换为真实版本列')
+assert.doesNotMatch(viewSource, /<el-table-column label="描述"/, '列表不应继续显示描述列')
+assert.match(viewSource, /current_version_nos/, '版本列应使用接口返回的真实当前版本号')
 assert.match(viewSource, /:multi-select="activeTab === 'task' && batchMode"/, '多选框只能在数据任务的批量模式显示')
 assert.match(viewSource, /:show-batch-bar="false"/, '数据任务应使用搜索框旁的批量删除入口')
 assert.match(viewSource, /<template #toolbar-extra>[\s\S]*?批量删除[\s\S]*?v-if="selectedCount > 0"[\s\S]*?删除（\{\{ selectedCount \}\}）/, '批量删除按钮没有按选择状态切换')
--- a/frontend/src/api/modules/dataset.ts
+++ b/frontend/src/api/modules/dataset.ts
@@ -1,5 +1,10 @@
 import { get, post, put, del } from '../request'
-import type { DatasetItem, DatasetVersion, DatasetVersionList } from '@/types'
+import type {
+  DatasetItem,
+  DatasetRecordSource,
+  DatasetVersion,
+  DatasetVersionList,
+} from '@/types'

 /** 数据集列表 */
 export const getDatasetList = () => get<DatasetItem[]>('/dataset-manage')
@@ -31,6 +36,12 @@ export const uploadDatasetFiles = (datasetId: string | number, files: File[]) =>
 export const previewDatasetFile = (fileId: string | number) =>
  get<{ content: string }>(`/dataset-manage/preview/${encodeURIComponent(fileId)}`)

+/** 获取发布样本所关联的原文参照 */
+export const getDatasetFileRecordSources = (fileId: string | number) =>
+  get<{ items: DatasetRecordSource[] }>(
+    `/dataset-manage/records/${encodeURIComponent(fileId)}/sources`,
+  )
+
 /** 数据集文件版本列表 */
 export const getDatasetFileVersions = (fileId: string | number) =>
  get<DatasetVersionList>(`/dataset-manage/versions/${encodeURIComponent(fileId)}`)
--- a/frontend/src/types/index.ts
+++ b/frontend/src/types/index.ts
@@ -63,6 +63,11 @@ export interface DatasetFile {
  id?: string
  name: string
  size?: string | number
+  size_bytes?: number
+  active_version_id?: string | null
+  current_version_id?: string | null
+  current_version_no?: number | null
+  version_count?: number
  record_count?: number
  split?: 'train' | 'validation' | 'test' | null
  [k: string]: any
@@ -76,13 +81,26 @@ export interface DatasetItem {
  source?: DatasetSource
  task_id?: string | number
  size?: string | number
+  size_bytes?: number
  count?: number
  description?: string
  create_time?: string
  files?: DatasetFile[]
+  current_version_no?: number | null
+  current_version_nos?: number[]
+  version_count?: number
  split_counts?: { train: number; validation: number; test: number }
 }

+export interface DatasetRecordSource {
+  line_no: number
+  instruction: string
+  input: string
+  output: string
+  source_text: string
+  preprocessed: boolean
+}
+
 export interface DatasetVersion {
  id: string
  version: number
--- a/frontend/src/utils/fileSize.ts
+++ b/frontend/src/utils/fileSize.ts
@@ -0,0 +1,26 @@
+export function parseSizeBytes(value?: string | number | null) {
+  if (typeof value === 'number') return Number.isFinite(value) ? Math.max(0, value) : 0
+  const match = String(value || '').trim().match(/^(\d+(?:\.\d+)?)\s*(B|KB|MB|GB|TB)?$/i)
+  if (!match) return 0
+  const units: Record<string, number> = {
+    B: 1,
+    KB: 1024,
+    MB: 1024 ** 2,
+    GB: 1024 ** 3,
+    TB: 1024 ** 4,
+  }
+  return Number(match[1]) * units[(match[2] || 'B').toUpperCase()]
+}
+
+/** 列表统一使用 MB，非零的小文件至少展示为 0.01 MB，避免误显示为零。 */
+export function formatMegabytes(
+  sizeBytes?: number | null,
+  legacySize?: string | number | null,
+) {
+  const numericSize = Number(sizeBytes)
+  const bytes = Number.isFinite(numericSize) && numericSize > 0
+    ? numericSize
+    : parseSizeBytes(legacySize)
+  if (bytes <= 0) return '0.00 MB'
+  return `${Math.max(bytes / 1024 ** 2, 0.01).toFixed(2)} MB`
+}
--- a/frontend/src/views/dataset/DatasetListView.vue
+++ b/frontend/src/views/dataset/DatasetListView.vue
@@ -5,6 +5,7 @@ import { ElMessage, ElMessageBox } from 'element-plus'
 import DataTablePage from '@/components/DataTablePage.vue'
 import { getDatasetList, deleteDataset, downloadDatasetUrl } from '@/api/modules/dataset'
 import { DATASET_TYPE_MAP, STORAGE_MAP } from '@/constants'
+import { formatMegabytes } from '@/utils/fileSize'
 import type { DatasetItem, DatasetSource } from '@/types'

 const router = useRouter()
@@ -107,6 +108,36 @@ function handleDownload(row: any) {
  window.open(downloadDatasetUrl(row.id), '_blank')
 }

+function formatSizeMb(row: DatasetItem) {
+  return formatMegabytes(row.size_bytes, row.size)
+}
+
+function currentVersionNumbers(row: DatasetItem) {
+  const values = row.current_version_nos?.length
+    ? row.current_version_nos
+    : [
+        row.current_version_no,
+        ...(row.files || []).map((file) => file.current_version_no),
+      ]
+  return [...new Set(
+    values
+      .map((value) => Number(value))
+      .filter((value) => Number.isInteger(value) && value > 0),
+  )].sort((a, b) => a - b)
+}
+
+function versionLabel(row: DatasetItem) {
+  const versions = currentVersionNumbers(row)
+  return versions.length ? versions.map((value) => `V${value}`).join(' / ') : '-'
+}
+
+function versionDescription(row: DatasetItem) {
+  const fileCount = row.files?.length || 0
+  const historyCount = Number(row.version_count || 0)
+  if (!fileCount) return '当前数据集没有文件版本'
+  return `${fileCount} 个文件，共 ${historyCount || fileCount} 个版本记录`
+}
+
 watch(activeTab, exitBatchMode)
 onMounted(loadData)
 </script>
@@ -118,7 +149,7 @@ onMounted(loadData)
    :data="filteredDataList"
    :loading="loading"
    searchable
-    :search-fields="['task_id', 'name', 'description']"
+    :search-fields="['task_id', 'name']"
    :multi-select="activeTab === 'task' && batchMode"
    :show-batch-bar="false"
    :create-text="activeTab === 'upload' ? '上传数据集' : ''"
@@ -181,16 +212,26 @@ onMounted(loadData)
          </el-tag>
        </template>
      </el-table-column>
-      <el-table-column label="大小" align="center" width="100">
+      <el-table-column label="大小" align="center" width="120">
        <template #default="{ row }">
-          {{ row.size && row.size !== '0 B' && row.size !== '0' ? row.size : '-' }}
+          <span class="numeric-cell">{{ formatSizeMb(row as DatasetItem) }}</span>
        </template>
      </el-table-column>
      <el-table-column label="数据条数" align="center" width="100">
        <template #default="{ row }">{{ row.count || 0 }}</template>
      </el-table-column>
-      <el-table-column label="描述" align="center" show-overflow-tooltip>
-        <template #default="{ row }">{{ row.description || '-' }}</template>
+      <el-table-column label="版本" align="center" width="120">
+        <template #default="{ row }">
+          <el-tooltip
+            :content="versionDescription(row as DatasetItem)"
+            placement="top"
+            :show-after="350"
+          >
+            <el-tag size="small" type="info" effect="plain">
+              {{ versionLabel(row as DatasetItem) }}
+            </el-tag>
+          </el-tooltip>
+        </template>
      </el-table-column>
      <el-table-column label="创建时间" align="center" width="180">
        <template #default="{ row }">
@@ -257,4 +298,9 @@ onMounted(loadData)
  }
 }

+.numeric-cell {
+  font-variant-numeric: tabular-nums;
+  white-space: nowrap;
+}
+
 </style>
--- a/frontend/src/views/dataset/DatasetPreviewView.vue
+++ b/frontend/src/views/dataset/DatasetPreviewView.vue
@@ -10,18 +10,27 @@ import {
  deleteDatasetFileVersion,
  downloadFileUrl,
  getDataset,
+  getDatasetFileRecordSources,
  getDatasetFileVersionContent,
  getDatasetFileVersions,
 } from '@/api/modules/dataset'
 import { DATASET_TYPE_MAP, STORAGE_MAP } from '@/constants'
-import { parseDatasetRecords, updateDatasetRecord } from './datasetRecords'
-import type { DatasetRecord } from './datasetRecords'
+import {
+  attachDatasetRecordSources,
+  parseDatasetRecords,
+} from './datasetRecords'
 import DatasetVersionBar from './preview/DatasetVersionBar.vue'
 import DatasetRecordTable from './preview/DatasetRecordTable.vue'
 import DatasetRecordEditorDialog from './preview/DatasetRecordEditorDialog.vue'
 import DatasetRawPreview from './preview/DatasetRawPreview.vue'
-import type { EditField } from './preview/types'
-import type { DatasetFile, DatasetItem, DatasetVersion } from '@/types'
+import { useDatasetRecordEditor } from './preview/useDatasetRecordEditor'
+import { formatMegabytes } from '@/utils/fileSize'
+import type {
+  DatasetFile,
+  DatasetItem,
+  DatasetRecordSource,
+  DatasetVersion,
+} from '@/types'

 const route = useRoute()
 const datasetId = route.params.id as string
@@ -34,14 +43,9 @@ const selectedFileId = ref('')
 const previewContent = ref('')
 const baseVersionContent = ref('')
 const previewError = ref('')
-const editorVisible = ref(false)
-const editingRecord = ref<DatasetRecord | null>(null)
-const editFields = ref<EditField[]>([])
-const rawDraft = ref('')
-const originalDraft = ref('')
-const savingRecord = ref(false)
 const savingVersion = ref(false)
 const versions = ref<DatasetVersion[]>([])
+const recordSources = ref<DatasetRecordSource[]>([])
 const selectedVersionId = ref('')
 const activeVersionId = ref('')
 const loadedVersionId = ref('')
@@ -56,15 +60,13 @@ const selectedFile = computed(() => files.value.find((file) => fileKey(file) ===
 const totalLines = computed(() => previewContent.value ? previewContent.value.split('\n').length : 0)
 const recordResult = computed(() => parseDatasetRecords(previewContent.value, selectedFile.value?.name || ''))
 const isRecordFile = computed(() => recordResult.value.supported)
-const records = computed(() => recordResult.value.records)
+const records = computed(() => attachDatasetRecordSources(
+  recordResult.value.records,
+  recordSources.value,
+))
 const viewedVersion = computed(() => versions.value.find((item) => item.id === loadedVersionId.value))
 const isViewingActiveVersion = computed(() => Boolean(loadedVersionId.value) && loadedVersionId.value === activeVersionId.value)
-const currentDraft = computed(() => editingRecord.value?.kind === 'object'
-  ? JSON.stringify(editFields.value)
-  : rawDraft.value)
-const hasEditorDraftChanges = computed(() => editorVisible.value && currentDraft.value !== originalDraft.value)
 const hasPendingVersionChanges = computed(() => previewContent.value !== baseVersionContent.value)
-const hasUnsavedChanges = computed(() => hasEditorDraftChanges.value || hasPendingVersionChanges.value)
 const datasetType = computed(() => {
  const value = String(dataset.value?.type || '').toLowerCase()
  return DATASET_TYPE_MAP[value] || dataset.value?.type || '-'
@@ -76,11 +78,28 @@ const storageType = computed(() => {
 const createdAt = computed(() => dataset.value?.create_time
  ? new Date(dataset.value.create_time).toLocaleString('zh-CN')
  : '-')
+const datasetSize = computed(() => formatMegabytes(
+  dataset.value?.size_bytes,
+  dataset.value?.size,
+))
+const selectedFileSize = computed(() => formatMegabytes(
+  selectedFile.value?.size_bytes,
+  selectedFile.value?.size,
+))

 function fileKey(file: DatasetFile) {
  return String(file.id || file.name)
 }

+const {
+  editorVisible, editingRecord, editFields, rawDraft, savingRecord,
+  hasEditorDraftChanges, openRecordEditor, resetRecordEditor, updateEditField,
+  closeRecordEditor, saveRecord, deleteRecord,
+} = useDatasetRecordEditor(selectedFile, isViewingActiveVersion, previewContent)
+const hasUnsavedChanges = computed(
+  () => hasEditorDraftChanges.value || hasPendingVersionChanges.value,
+)
+
 async function loadDataset() {
  loading.value = true
  try {
@@ -101,7 +120,10 @@ async function loadVersions(file: DatasetFile) {
  const requestId = ++versionRequestId
  try {
    const result = await getDatasetFileVersions(fileKey(file))
-    const activeContent = await getDatasetFileVersionContent(fileKey(file), result.active_version_id)
+    const [activeContent, sources] = await Promise.all([
+      getDatasetFileVersionContent(fileKey(file), result.active_version_id),
+      getDatasetFileRecordSources(fileKey(file)),
+    ])
    if (requestId !== versionRequestId) return
    versions.value = result.versions
    nextVersionNumber.value = result.next_version_number
@@ -111,9 +133,13 @@ async function loadVersions(file: DatasetFile) {
    loadedVersionId.value = result.active_version_id
    previewContent.value = activeContent.content
    baseVersionContent.value = activeContent.content
+    recordSources.value = sources.items || []
    resetRecordEditor()
  } catch {
-    if (requestId === versionRequestId) previewError.value = '版本内容加载失败，请稍后重试'
+    if (requestId === versionRequestId) {
+      recordSources.value = []
+      previewError.value = '版本内容加载失败，请稍后重试'
+    }
  } finally {
    if (requestId === versionRequestId) previewLoading.value = false
  }
@@ -221,40 +247,6 @@ async function deleteViewedVersion() {
  }
 }

-function openRecordEditor(record: DatasetRecord) {
-  if (!isViewingActiveVersion.value) {
-    ElMessage.warning('历史版本为只读，请先设为当前版本')
-    return
-  }
-  editingRecord.value = record
-  if (record.kind === 'object') {
-    editFields.value = Object.entries(record.value as Record<string, unknown>).map(([key, value]) => ({
-      key,
-      value: typeof value === 'string' ? value : JSON.stringify(value, null, 2),
-      isJson: typeof value !== 'string',
-    }))
-    rawDraft.value = ''
-  } else {
-    editFields.value = []
-    rawDraft.value = record.raw
-  }
-  editorVisible.value = true
-  originalDraft.value = record.kind === 'object' ? JSON.stringify(editFields.value) : rawDraft.value
-}
-
-function resetRecordEditor() {
-  editorVisible.value = false
-  editingRecord.value = null
-  editFields.value = []
-  rawDraft.value = ''
-  originalDraft.value = ''
-}
-
-function updateEditField(index: number, value: string) {
-  const field = editFields.value[index]
-  if (field) field.value = value
-}
-
 async function confirmDiscardChanges(message = '当前存在尚未保存为版本的修改，放弃后将无法恢复。') {
  if (!hasUnsavedChanges.value) return true
  try {
@@ -269,63 +261,6 @@ async function confirmDiscardChanges(message = '当前存在尚未保存为版
  }
 }

-async function confirmEditorDiscardChanges() {
-  if (!hasEditorDraftChanges.value) return true
-  try {
-    await ElMessageBox.confirm('当前样本存在未暂存修改，放弃后将无法恢复。', '确认放弃修改？', {
-      type: 'warning',
-      confirmButtonText: '放弃修改',
-      cancelButtonText: '继续编辑',
-    })
-    return true
-  } catch {
-    return false
-  }
-}
-
-async function closeRecordEditor(done?: () => void) {
-  if (savingRecord.value) return
-  if (!await confirmEditorDiscardChanges()) return
-  resetRecordEditor()
-  done?.()
-}
-
-function buildNextRecordValue() {
-  if (editingRecord.value?.kind !== 'object') return JSON.parse(rawDraft.value)
-  const nextValue: Record<string, unknown> = {}
-  for (const field of editFields.value) {
-    nextValue[field.key] = field.isJson ? JSON.parse(field.value) : field.value
-  }
-  return nextValue
-}
-
-async function saveRecord() {
-  if (!selectedFile.value || !editingRecord.value || savingRecord.value) return
-  const record = editingRecord.value
-  let nextValue: unknown
-  try {
-    nextValue = buildNextRecordValue()
-  } catch (error) {
-    ElMessage.error(`JSON 格式校验失败：${error instanceof Error ? error.message : '内容格式不正确'}`)
-    return
-  }
-
-  savingRecord.value = true
-  try {
-    const nextContent = updateDatasetRecord(
-      previewContent.value,
-      selectedFile.value.name,
-      record.sourceIndex,
-      nextValue,
-    )
-    previewContent.value = nextContent
-    resetRecordEditor()
-    ElMessage.success(`第 ${record.displayIndex} 条修改已暂存`)
-  } finally {
-    savingRecord.value = false
-  }
-}
-
 async function saveVersion() {
  if (
    !selectedFile.value
@@ -396,7 +331,7 @@ onBeforeUnmount(() => window.removeEventListener('beforeunload', handleBeforeUnl

    <section class="dataset-overview" aria-label="数据集概览">
      <div class="overview-item"><span>数据条数</span><strong>{{ (dataset?.count || 0).toLocaleString() }}</strong></div>
-      <div class="overview-item"><span>数据大小</span><strong>{{ dataset?.size || '-' }}</strong></div>
+      <div class="overview-item"><span>数据大小</span><strong>{{ datasetSize }}</strong></div>
      <div class="overview-item overview-item-wide"><span>创建时间</span><strong>{{ createdAt }}</strong></div>
    </section>

@@ -433,7 +368,7 @@ onBeforeUnmount(() => window.removeEventListener('beforeunload', handleBeforeUnl
            <div>
              <h2>{{ selectedFile?.name }}</h2>
              <p>
-                {{ selectedFile?.size || '未知大小' }} <span aria-hidden="true">·</span>
+                {{ selectedFileSize }} <span aria-hidden="true">·</span>
                <template v-if="isRecordFile">{{ records.length.toLocaleString() }} 条记录</template>
                <template v-else>{{ totalLines.toLocaleString() }} 行</template>
              </p>
@@ -470,6 +405,7 @@ onBeforeUnmount(() => window.removeEventListener('beforeunload', handleBeforeUnl
            :is-viewing-active-version="isViewingActiveVersion"
            :version-id="loadedVersionId"
            @edit="openRecordEditor"
+            @delete="deleteRecord"
          />

          <DatasetRawPreview
--- a/frontend/src/views/dataset/datasetRecords.ts
+++ b/frontend/src/views/dataset/datasetRecords.ts
@@ -1,3 +1,5 @@
+import type { DatasetRecordSource } from '@/types'
+
 export type DatasetRecordKind = 'object' | 'primitive' | 'invalid'

 export interface DatasetRecord {
@@ -6,6 +8,8 @@ export interface DatasetRecord {
  raw: string
  kind: DatasetRecordKind
  value: unknown
+  sourceText?: string
+  sourceWasPreprocessed?: boolean
  error?: string
 }

@@ -15,6 +19,35 @@ export interface DatasetRecordResult {
  error?: string
 }

+/** 通过内容优先、物理行兜底，把发布样本重新关联到切片原文。 */
+export function attachDatasetRecordSources(
+  records: DatasetRecord[],
+  sources: DatasetRecordSource[],
+) {
+  const used = new Set<number>()
+  return records.map((record) => {
+    const value = record.kind === 'object'
+      ? record.value as Record<string, unknown>
+      : {}
+    let sourceIndex = sources.findIndex((source, index) => (
+      !used.has(index)
+      && source.instruction === String(value.instruction || '')
+      && source.input === String(value.input || '')
+      && source.output === String(value.output || '')
+    ))
+    if (sourceIndex < 0 && sources[record.sourceIndex] && !used.has(record.sourceIndex)) {
+      sourceIndex = record.sourceIndex
+    }
+    if (sourceIndex < 0) return record
+    used.add(sourceIndex)
+    return {
+      ...record,
+      sourceText: sources[sourceIndex]?.source_text || '',
+      sourceWasPreprocessed: sources[sourceIndex]?.preprocessed || false,
+    }
+  })
+}
+
 function isObjectRecord(value: unknown): value is Record<string, unknown> {
  return typeof value === 'object' && value !== null && !Array.isArray(value)
 }
@@ -119,3 +152,33 @@ export function updateDatasetRecord(

  throw new Error('当前文件不支持逐条编辑')
 }
+
+/** 删除目标样本；删除只修改当前草稿，仍需由调用方保存为新版本。 */
+export function deleteDatasetRecord(
+  content: string,
+  fileName: string,
+  sourceIndex: number,
+) {
+  const normalizedName = fileName.toLowerCase()
+  if (normalizedName.endsWith('.jsonl')) {
+    const lines = content.split('\n')
+    if (sourceIndex < 0 || sourceIndex >= lines.length || !lines[sourceIndex]?.trim()) {
+      throw new Error('目标样本不存在')
+    }
+    lines.splice(sourceIndex, 1)
+    return lines.join('\n')
+  }
+
+  if (normalizedName.endsWith('.json')) {
+    const root: unknown = JSON.parse(content)
+    if (Array.isArray(root)) {
+      if (sourceIndex < 0 || sourceIndex >= root.length) throw new Error('目标样本不存在')
+      root.splice(sourceIndex, 1)
+      return JSON.stringify(root, null, 2)
+    }
+    if (sourceIndex !== 0) throw new Error('目标样本不存在')
+    return '[]'
+  }
+
+  throw new Error('当前文件不支持逐条删除')
+}
--- a/frontend/src/views/dataset/preview/DatasetRecordTable.vue
+++ b/frontend/src/views/dataset/preview/DatasetRecordTable.vue
@@ -10,6 +10,7 @@ const props = defineProps<{

 const emit = defineEmits<{
  edit: [record: DatasetRecord]
+  delete: [record: DatasetRecord]
 }>()

 const pageSize = 10
@@ -18,9 +19,18 @@ const currentPage = ref(1)

 const allTableFieldKeys = computed(() => {
  const keys = new Set<string>()
+  const sourceFieldKeys = new Set([
+    'source_text',
+    'source_content',
+    'original_text',
+    'original',
+    'source',
+  ])
  props.records.forEach((record) => {
    if (record.kind !== 'object') return
-    Object.keys(record.value as Record<string, unknown>).forEach((key) => keys.add(key))
+    Object.keys(record.value as Record<string, unknown>)
+      .filter((key) => !sourceFieldKeys.has(key))
+      .forEach((key) => keys.add(key))
  })
  const preferred = ['instruction', 'question', 'prompt', 'input', 'context', 'output', 'answer', 'response']
  return [...keys].sort((a, b) => {
@@ -69,6 +79,16 @@ function recordFieldValue(record: DatasetRecord, key: string) {
  return Object.prototype.hasOwnProperty.call(data, key) ? formatFieldValue(data[key]) : '-'
 }

+function recordSourceText(record: DatasetRecord) {
+  if (record.sourceText?.trim()) return record.sourceText
+  if (record.kind !== 'object') return '无关联原文'
+  const data = record.value as Record<string, unknown>
+  for (const key of ['source_text', 'source_content', 'original_text', 'original', 'source']) {
+    if (typeof data[key] === 'string' && data[key].trim()) return data[key]
+  }
+  return '无关联原文'
+}
+
 function extraFieldValue(record: DatasetRecord) {
  if (record.kind !== 'object') return '-'
  const data = record.value as Record<string, unknown>
@@ -144,6 +164,22 @@ function recordRowClassName({ row }: { row: unknown }) {
            </el-tooltip>
          </template>
        </el-table-column>
+        <el-table-column label="原文参照" min-width="280">
+          <template #default="{ row }">
+            <el-tooltip
+              :content="recordSourceText(asDatasetRecord(row))"
+              placement="top"
+              :show-after="400"
+            >
+              <div
+                class="table-cell-text source-reference-cell"
+                :class="{ 'is-empty': recordSourceText(asDatasetRecord(row)) === '无关联原文' }"
+              >
+                {{ recordSourceText(asDatasetRecord(row)) }}
+              </div>
+            </el-tooltip>
+          </template>
+        </el-table-column>
        <el-table-column v-if="hasExtraTableFields" label="其他字段" min-width="240">
          <template #default="{ row }">
            <el-tooltip
@@ -166,19 +202,32 @@ function recordRowClassName({ row }: { row: unknown }) {
            <el-tag v-else size="small" type="success">正常</el-tag>
          </template>
        </el-table-column>
-        <el-table-column label="操作" width="96" align="center" fixed="right">
+        <el-table-column label="操作" width="156" align="center" fixed="right">
          <template #default="{ row }">
-            <el-button
-              type="primary"
-              link
-              :disabled="!isViewingActiveVersion"
-              :title="isViewingActiveVersion ? '' : '历史版本为只读，请先设为当前版本'"
-              :aria-label="`${asDatasetRecord(row).kind === 'invalid' ? '修复' : '编辑'}第 ${asDatasetRecord(row).displayIndex} 条数据`"
-              @click="emit('edit', asDatasetRecord(row))"
-            >
-              <i :class="asDatasetRecord(row).kind === 'invalid' ? 'fa fa-wrench' : 'fa fa-pencil'" />
-              {{ asDatasetRecord(row).kind === 'invalid' ? '修复' : '编辑' }}
-            </el-button>
+            <div class="record-actions">
+              <el-button
+                type="primary"
+                link
+                :disabled="!isViewingActiveVersion"
+                :title="isViewingActiveVersion ? '' : '历史版本为只读，请先设为当前版本'"
+                :aria-label="`${asDatasetRecord(row).kind === 'invalid' ? '修复' : '编辑'}第 ${asDatasetRecord(row).displayIndex} 条数据`"
+                @click="emit('edit', asDatasetRecord(row))"
+              >
+                <i :class="asDatasetRecord(row).kind === 'invalid' ? 'fa fa-wrench' : 'fa fa-pencil'" />
+                {{ asDatasetRecord(row).kind === 'invalid' ? '修复' : '编辑' }}
+              </el-button>
+              <el-button
+                type="danger"
+                link
+                :disabled="!isViewingActiveVersion"
+                :title="isViewingActiveVersion ? '' : '历史版本为只读，请先设为当前版本'"
+                :aria-label="`删除第 ${asDatasetRecord(row).displayIndex} 条数据`"
+                @click="emit('delete', asDatasetRecord(row))"
+              >
+                <i class="fa fa-trash-o" />
+                删除
+              </el-button>
+            </div>
          </template>
        </el-table-column>
      </el-table>
@@ -287,6 +336,21 @@ function recordRowClassName({ row }: { row: unknown }) {
  -webkit-line-clamp: 2;
 }

+.source-reference-cell {
+  color: #344054;
+}
+
+.source-reference-cell.is-empty {
+  color: #98a2b3;
+}
+
+.record-actions {
+  display: inline-flex;
+  align-items: center;
+  justify-content: center;
+  white-space: nowrap;
+}
+
 .record-table .el-button i {
  margin-right: 5px;
 }
--- a/frontend/src/views/dataset/preview/useDatasetRecordEditor.ts
+++ b/frontend/src/views/dataset/preview/useDatasetRecordEditor.ts
@@ -0,0 +1,174 @@
+import { computed, ref } from 'vue'
+import type { ComputedRef, Ref } from 'vue'
+import { ElMessage, ElMessageBox } from 'element-plus'
+import {
+  deleteDatasetRecord,
+  updateDatasetRecord,
+} from '../datasetRecords'
+import type { DatasetRecord } from '../datasetRecords'
+import type { DatasetFile } from '@/types'
+import type { EditField } from './types'
+
+export function useDatasetRecordEditor(
+  selectedFile: ComputedRef<DatasetFile | undefined>,
+  isViewingActiveVersion: ComputedRef<boolean>,
+  previewContent: Ref<string>,
+) {
+  const editorVisible = ref(false)
+  const editingRecord = ref<DatasetRecord | null>(null)
+  const editFields = ref<EditField[]>([])
+  const rawDraft = ref('')
+  const originalDraft = ref('')
+  const savingRecord = ref(false)
+  const currentDraft = computed(() => editingRecord.value?.kind === 'object'
+    ? JSON.stringify(editFields.value)
+    : rawDraft.value)
+  const hasEditorDraftChanges = computed(
+    () => editorVisible.value && currentDraft.value !== originalDraft.value,
+  )
+
+  function resetRecordEditor() {
+    editorVisible.value = false
+    editingRecord.value = null
+    editFields.value = []
+    rawDraft.value = ''
+    originalDraft.value = ''
+  }
+
+  function openRecordEditor(record: DatasetRecord) {
+    if (!isViewingActiveVersion.value) {
+      ElMessage.warning('历史版本为只读，请先设为当前版本')
+      return
+    }
+    editingRecord.value = record
+    if (record.kind === 'object') {
+      editFields.value = Object.entries(record.value as Record<string, unknown>).map(
+        ([key, value]) => ({
+          key,
+          value: typeof value === 'string' ? value : JSON.stringify(value, null, 2),
+          isJson: typeof value !== 'string',
+        }),
+      )
+      rawDraft.value = ''
+    } else {
+      editFields.value = []
+      rawDraft.value = record.raw
+    }
+    editorVisible.value = true
+    originalDraft.value = record.kind === 'object'
+      ? JSON.stringify(editFields.value)
+      : rawDraft.value
+  }
+
+  function updateEditField(index: number, value: string) {
+    const field = editFields.value[index]
+    if (field) field.value = value
+  }
+
+  async function confirmEditorDiscardChanges() {
+    if (!hasEditorDraftChanges.value) return true
+    try {
+      await ElMessageBox.confirm(
+        '当前样本存在未暂存修改，放弃后将无法恢复。',
+        '确认放弃修改？',
+        {
+          type: 'warning',
+          confirmButtonText: '放弃修改',
+          cancelButtonText: '继续编辑',
+        },
+      )
+      return true
+    } catch {
+      return false
+    }
+  }
+
+  async function closeRecordEditor(done?: () => void) {
+    if (savingRecord.value || !await confirmEditorDiscardChanges()) return
+    resetRecordEditor()
+    done?.()
+  }
+
+  function buildNextRecordValue() {
+    if (editingRecord.value?.kind !== 'object') return JSON.parse(rawDraft.value)
+    const nextValue: Record<string, unknown> = {}
+    for (const field of editFields.value) {
+      nextValue[field.key] = field.isJson ? JSON.parse(field.value) : field.value
+    }
+    return nextValue
+  }
+
+  async function saveRecord() {
+    if (!selectedFile.value || !editingRecord.value || savingRecord.value) return
+    const record = editingRecord.value
+    let nextValue: unknown
+    try {
+      nextValue = buildNextRecordValue()
+    } catch (error) {
+      ElMessage.error(`JSON 格式校验失败：${error instanceof Error ? error.message : '内容格式不正确'}`)
+      return
+    }
+
+    savingRecord.value = true
+    try {
+      previewContent.value = updateDatasetRecord(
+        previewContent.value,
+        selectedFile.value.name,
+        record.sourceIndex,
+        nextValue,
+      )
+      resetRecordEditor()
+      ElMessage.success(`第 ${record.displayIndex} 条修改已暂存`)
+    } finally {
+      savingRecord.value = false
+    }
+  }
+
+  async function deleteRecord(record: DatasetRecord) {
+    if (!selectedFile.value || !isViewingActiveVersion.value) {
+      ElMessage.warning('历史版本为只读，请先设为当前版本')
+      return
+    }
+    try {
+      await ElMessageBox.confirm(
+        '删除后将暂存到当前草稿，点击“保存版本”后正式生效。',
+        `删除第 ${record.displayIndex} 条数据？`,
+        {
+          type: 'warning',
+          confirmButtonText: '删除',
+          cancelButtonText: '取消',
+          confirmButtonClass: 'el-button--danger',
+        },
+      )
+    } catch {
+      return
+    }
+
+    try {
+      previewContent.value = deleteDatasetRecord(
+        previewContent.value,
+        selectedFile.value.name,
+        record.sourceIndex,
+      )
+      resetRecordEditor()
+      ElMessage.success(`第 ${record.displayIndex} 条已删除，请保存为新版本`)
+    } catch (error) {
+      ElMessage.error(error instanceof Error ? error.message : '删除失败')
+    }
+  }
+
+  return {
+    editorVisible,
+    editingRecord,
+    editFields,
+    rawDraft,
+    savingRecord,
+    hasEditorDraftChanges,
+    openRecordEditor,
+    resetRecordEditor,
+    updateEditField,
+    closeRecordEditor,
+    saveRecord,
+    deleteRecord,
+  }
+}
Author	SHA1	Message	Date
caoxiaozhu	e6bbb0bb49	feat(dataset): 补全样本原文与删除操作	2026-07-27 12:26:19 +08:00
caoxiaozhu	bce586697b	fix(dataset): 统一大小与版本元数据	2026-07-27 12:26:09 +08:00